TW201329785A - Interactive voice command editing system and method - Google Patents

Abstract

The present invention relates to an interactive voice command editing method. The method comprises: recording a new action and determining the new action whether it is same with one historical action from historical action information while comparing the new action with the historical action information. It establishes a new voice command and generates a new voice command feature vector. Afterward, verifying the new voice command feature vector according to a predetermined threshold value so as to store the new voice command.

Description

Interactive voice instruction construction system and method

The present invention relates to an interactive voice instruction construction method and system, and more particularly to an interactive voice instruction construction method and system that can generate a corresponding voice command according to a new action.

In recent years, the popularity of consumer electronics has increased in life, and users have increased significantly. In various applications, there are personal hand-held systems, tablet computers, and vehicle electronics and robots, electronic toys, etc. combined with daily life tools. However, the convenience of using these electronic products highlights the importance of the human-machine interface. The purpose of the present invention is to create a convenient human-computer interaction.

When people use these electronic products, the most intuitive thing is to operate or use voice commands. The operation of these electronic products requires a series of related actions, which often gives users the idea of resisting the use of electronic products. Or the voice command fixes the relevant voice command with the machine device, and the user does not easily modify it. The present invention expresses this series of actions in the most intuitive voice commands. Let the user associate new actions with their voice commands to increase the ease of operation.

In an embodiment, the present invention provides an interactive voice instruction construction method, including: recording a new action including determining a start of an action; and comparing the new action with historical action information to determine whether there is the same as the new action action. If it is a new action, a voice command is added, and the voice command is based on a preset threshold to verify whether the voice command needs to be generated.

In an embodiment, the present invention provides an interactive voice instruction construction system, including: an action start/end detection module to determine the start of a new action; and an action recording module for recording a new action; The library is configured to store historical action information and historical voice commands; and an action comparison module, receive the new action, and compare the new action with the historical action information to determine whether there is the same action as the new action. A voice command is added to the module to add a new voice command corresponding to the new action; a voice command verification module is used to verify the new voice command and the historical voice command with the threshold value, and whether the new voice command is verified.

In order to enable the reviewing committee to have a further understanding and understanding of the features, objects and functions of the present invention, the related detailed structure of the device of the present invention and the concept of the design are explained below so that the reviewing committee can understand the present invention. Features, detailed descriptions are set forth below: Figure 1 shows an interactive voice command construction system 1 in accordance with an embodiment of the present invention. The interactive voice construction system includes: an action start/end detection module 10, an action record module 11, a database 12, an action comparison module 13, a voice command addition module 14, and a voice command verification. Module 15. The action start/end detection module 10 determines the start and end of the action by establishing a motion rule table by sensing the signal of the device or the camera. The action recording module 11 can capture and record a new action by using a sensing device or a photographing device, and the new action can be an object action, such as an object action such as a vehicle or a toy, and the sensing device and the photographing device It may include a camera, a sensor, and the like. The database 12 can be used to store historical action information and historical voice commands. The action comparison module 13 can receive the new action and compare the new action with the historical action information to determine whether there is the same action as the new action. If there is no action of the new action in the historical action information, a voice is input according to the new action, so that the voice command adding module 14 creates a new voice command and generates a new voice command feature value. The voice command verification module 15 can preset a threshold according to one of the historical voice commands to verify the new voice command feature value. The threshold value can be established by setting an empirical value or by other algorithms. The new voice command feature value and the historical voice command feature value in the database 12 can be obtained by a log-Likelihood score difference, and the score difference is compared with the preset threshold value. And determining whether the new voice command is established or not, if the logarithmic similarity score difference between the new voice command feature value and the historical voice command feature value is greater than the preset threshold, storing the new voice command to the database 12, and the voice The command verification module 14 adds the new voice command feature value to the historical voice command database 12 as one of the historical voice command feature value log similarity score range values, as the basis for calculating the score gap for the next new voice command; Otherwise, if the logarithmic similarity score difference between the new voice command feature value and the historical voice command feature value is less than the preset threshold, the voice command verification module discards the new voice command or overwrites an original historical voice command. Corresponding action. In addition, if the historical action information has the same action as the new action, the action comparison module 14 discards the establishment of the new voice command or changes an original voice command. In addition, when the new action is not a directly alignable action and cannot be directly compared with the historical action information, an algorithm is further applied to identify the new action, and the algorithm may be a neural network algorithm, Decision tree algorithm or support vector machine (SVM) and so on. Neural network algorithms such as the foregoing include mechanical rote learning, learning by instruction, learning by analogy, and learning by induction.

FIG. 2 is a flowchart showing the action comparison of an interactive voice command construction method according to an embodiment of the present invention, and the action comparison process can be combined with the interactive voice command construction system of FIG. 1 and described. First, the application of the interactive voice construction method includes: (Step S201) the application action start/end detection module 10 and the motion recording module 11 capture and record a new action by using a sensing device or a photographing device. The new action may be an object action, such as an object or vehicle action, and the sensing device and the photographic device may include a camera, a sensor, or the like. (Step S202) Applying the action comparison module 13 to compare the new action with the historical action information stored in the database (step S203) to determine whether there is the same action as the new action, if the historical action information is The action is not the same as the new action, (step S204), the voice command adding module 14 is activated, and a voice is further input according to the new action to establish a new voice command and generate a new voice command feature value; otherwise, (Step S205) If the historical action information has the same action as the new action, the new voice command is abandoned and a new action is re-recorded or an original voice command is changed to activate the voice command adding module 14. In addition, when the new action is not a directly alignable action and cannot be directly compared with the historical action information, an algorithm is further applied to identify the new action, and the algorithm may be a neural network algorithm, Decision tree algorithm or support vector machine (SVM) and so on. Neural network algorithms such as the foregoing include mechanical rote learning, learning by instruction, learning by analogy, and learning by induction.

FIG. 3 is a flowchart of a verification voice instruction according to an interactive voice instruction construction method according to an embodiment of the present invention, and the flowchart of constructing a voice instruction may also be compared with the flowchart of FIG. 2 and the interactive voice of FIG. Construct a system to match and explain. After the voice command adding module 14 is activated, and a voice is further input according to the new action to establish a new voice command and a new voice command feature value is generated, (step S301), the new voice command can be verified. Next, (step S302), according to a log-Likelihood, comparing a difference between the logarithmic similarity score value of the historical voice command feature value and the new voice command feature value with a preset threshold value, Verifying the new voice command feature value, if the logarithmic similarity score difference between the new voice command feature value and the historical voice command feature value is greater than the preset threshold value, the new voice command feature value is used as a new voice command parameter, Step S303) and storing the new voice command to the database 12 as one of the basis for calculating the log similarity score of the next new voice command; otherwise, if the new voice command feature value is compared with the logarithmic similarity of the historical voice command feature value The score difference is less than the preset threshold value (step S304), the new voice command is overwritten or the action corresponding to an original historical voice command is overwritten to be stored in the database 12 (step S305) and the new voice command is re-entered. .

In addition, the voice command addition module 14 of the present invention can establish a speaker dependent voice command and a speaker independent voice command. The speaker-related voice command can establish the characteristics related to the user's voice. The most common algorithm is dynamic time warping (DTW). The DTW algorithm applies two similar sound files to the speech recognition to establish the voice command. Firstly, the eigenvalue parameter of the sound file is obtained, and then the voice command model parameter is established by the DTW algorithm. The speaker unrelated speech command can apply the speech acoustic model trained by the hidden markov model (HMM) algorithm, and the user records the audio file to make the model parameter adjustment to establish the voice command. The common identification method is the Viterbi re-estimation algorithm. The user first obtains the characteristic value parameter from the recorded command sound file, and then combines with the original HMM model to adjust the new voice. Instruction model parameters.

The interactive voice instruction construction system and method of the present invention can be applied to an embodiment of a toy car interactive voice command construction system. In this application, the user operates the toy car to establish a voice command, and when the toy car performs an "S"-shaped action, the action start/end detection module 10 and the action record module 11 in the system can be just the user. The sequence of actions performed is recorded, and then the action comparison module 13 is compared with the database 12 to determine whether it is a new action. If it is a new action, the system activates the voice command increase module 14 to perform a voice command. Input; otherwise, if the action is an old action, the system will ask the user whether to change the original voice command or abandon the establishment of the voice command to re-wait and record the occurrence of "new action". The action of the toy car can be indicated by the rotational position on each axle servo motor. When the new action is established, the system can prompt the user to record and input a voice command, and the user can input the voice command to be represented for the new action, and the voice command can be one or more voice files, such as "folk snake" Wait. At this time, the voice command adding module 14 converts the voice input by the user into a set of voice model parameters through the voice recognition algorithm, and then verifies through the voice verification module 14 and the voice command in the database 12. The voice command verification module 14 will perform the verification standard according to a threshold value. When the verification value is greater than the set threshold value, a new voice command will be generated, and the system will corresponding the new voice command to the new action. Later, when the user performs the manipulation of the S-shaped motion, the voice command can be directly input to speed up the processing efficiency.

FIG. 4 shows another embodiment of the construction of interactive voice instructions for applying image processing on a computer system. As shown in FIG. 4, in the embodiment, when the user wants to reduce and rotate the picture into picture two, the user will perform a zoom-out action on the picture, and then perform a rotation action to obtain the picture two and store it. At this time, the action comparison module 11 in the interactive voice command construction system compares the sequence of actions performed by the user with the database 12, and if found to be a new action, the user is prompted whether to add a voice command. The user can input the voice command to be represented for the new action. At this time, the voice command adding module 14 converts the voice input by the user into a set of voice models, and then passes through the voice verification module 15 and the data library 12. The voice command is compared. When the comparison result is a new voice command, the system corresponding the new voice command to the new action. When the user wants to perform the same processing on the image in the future, the voice command can be directly input to speed up the processing.

The invention can dynamically generate voice commands by the user's operation. When the user completes an action, the system will perform an action comparison verification from the action database. If the new action system starts, the request for adding a voice command will be initiated. At this point, the user immediately records the voice command through the recording device, and then the system generates a new voice command via the voice command verification mechanism.

The above description is only exemplary of the invention, and the scope of the invention is not limited thereto. That is to say, the equivalent changes and modifications made by the applicant in accordance with the scope of the patent application of the present invention should still fall within the scope of the patent of the present invention. I would like to ask your review committee to give a clear explanation and pray for it.

1. . . Interactive voice command construction system

10. . . Motion start/end detection module

11. . . Motion recording module

12. . . database

13. . . Motion comparison module

14. . . Voice command addition module

15. . . Voice command verification module

S201~S205. . . step

S301~S305. . . step

1 shows an interactive voice instruction construction system 1 in accordance with an embodiment of the present invention.

2 is a flow chart showing the operation of an interactive voice command construction method according to an embodiment of the present invention.

FIG. 3 shows a flow verification flowchart of an interactive voice instruction construction method according to an embodiment of the present invention.

FIG. 4 shows an embodiment of the construction of interactive voice instructions for applying image processing on a computer system.

S201~S205. . . step

Claims (15)

An interactive voice instruction construction method includes: recording a new action; and comparing the new action with historical action information to determine whether there is the same action as the new action. The method for constructing an interactive voice instruction as described in claim 1, further comprising: if the historical action information does not have the same action of the new action, inputting a voice according to the new action to establish a new voice command. And generate a new voice command feature value. The method for constructing an interactive voice instruction as described in claim 2, further comprising: presetting a threshold according to one of the historical voice commands to verify the new voice command feature value. The method for constructing an interactive voice instruction as described in claim 3, further comprising: if the logarithmic similarity score difference between the new voice command feature value and the historical voice command feature value is greater than the preset threshold value, storing the new The voice command, and the new voice command feature value is used as one of the ranges of the logarithmic similarity score difference for the next new voice command. The method for constructing an interactive voice instruction as described in claim 3, further comprising: if the logarithmic similarity score difference between the new voice command feature value and the historical voice command feature value is less than the preset threshold value, the new The voice command or the action corresponding to an original historical voice command. The method for constructing an interactive voice instruction as described in claim 2, further comprising: if the historical action information has the same action as the new action, abandoning the establishment of the new voice command or changing an original voice command. . The interactive voice instruction construction method according to claim 1, wherein when the new action cannot directly compare with the historical action information, an algorithm is further applied to identify the new action. For example, the interactive voice instruction construction method described in claim 7 is wherein the algorithm is a neural network algorithm, a decision tree algorithm or a support vector machine. An interactive voice command construction system includes: an action start/end detection module for determining the start of a new action; an action record module for recording a new action; and a database for storing The historical action information and the historical voice command; and an action comparison module receive the new action and compare the new action with the historical action information to determine whether there is the same action as the new action. The interactive voice instruction construction system of claim 9, further comprising: a voice command adding module, if the historical action information does not have the same action of the new action, inputting a voice according to the new action So that the voice command adding module creates a new voice command and generates a new voice command feature value; and a voice command verification module, according to a preset threshold value and a historical voice command feature value and a new voice command feature value The difference between the logarithmic similarity scores to verify the new voice instruction feature value. The interactive voice instruction construction system according to claim 10, wherein if the logarithmic similarity score difference between the new voice command feature value and the historical voice command feature value is greater than the preset threshold, the new voice command is stored. To the database, and the voice command verification module uses the new voice command feature value as one of the ranges of the log similarity score difference of the next new voice command. The interactive voice instruction construction system according to claim 10, wherein if the logarithmic similarity score difference between the new voice command feature value and the historical voice command feature value is less than the preset threshold, the voice command verification mode The group abandons the new voice command or overrides the action corresponding to an original historical voice command. The interactive voice instruction construction system according to claim 9, wherein if the historical action information has the same action as the new action, the action comparison module gives up the establishment of the new voice command or changes an original Some voice commands. The interactive voice instruction construction system according to claim 9, wherein when the new action cannot directly compare with the historical action information, an algorithm is further applied to identify the new action. For example, the interactive voice instruction construction system described in claim 14 is wherein the algorithm is a neural network algorithm, a decision tree algorithm or a support vector machine.